Basic data description procedures

A down-to-earth approach to this problem would list the chain of shared activities which feed into most R&D in the field, for example:

• Transcription: `canonical' (orthographic, phonemic, prosodic, paralinguistic) and `detailed'; often by semi-automatic methods.
• Annotation (labelling): Alignment of transcription symbols with points or intervals in the speech signal, using signal processing and display software such as ESPS/waves+, or automatic alignment software.
• Normalisation: the transcription is normalised (parsed, corrected, re-formatted) for further processing.
• Lexicon extraction: the basic units (generally fully-inflected words) in the lexicon are extracted from the corpus to serve as the basis for further processing.
• Statistical training of speech recognisers using the lexicon and the labelled data.
• Construction of statistical language models using the lexicon and the transcription corpus.
• Development of linguistic lexica for parsing.
• ...